Leveraging deep learning techniques to improvep300-based brain computerinterfaces
Başlık çevirisi mevcut değil.
- Tez No: 720978
- Danışmanlar: PROF. ALESSANDRA PEDROCCHI
- Tez Türü: Yüksek Lisans
- Konular: Biyomühendislik, Biyoteknoloji, Tıbbi Biyoloji, Bioengineering, Biotechnology, Medical Biology
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2021
- Dil: İngilizce
- Üniversite: Polıtecnıco Dı Mılano
- Enstitü: Yurtdışı Enstitü
- Ana Bilim Dalı: Belirtilmemiş.
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: Belirtilmemiş.
Özet
Özet yok.
Özet (Çeviri)
Brain-Computer Interface (BCI) has become a promising technology to carry out training sessions with Autism Spectrum Disorder (ASD) subjects to overcome their joint-attention deficit. The functioning is based on the successful classification of the P300 waveform from EEG recordings, an event-related potential elicited by rare stimuli in visual-oddball paradigms, with a latency of 300 ms after the onset of a stimulus. In this thesis work, we used deep learning techniques, namely convolutional neural networks (CNNs), to improve P300-based BCIs. We utilized a state-of-the-art classifier, named BCIAUT, which had proven success on a P300 data set acquired on ASD subjects, called BCIAUT-P300 data set. We also proposed a novel BCI classifier, called P3CNET, that improved P300 classification accuracy performances shown by BCIAUT. For the pre-processing of the data, we pursued a grid-search approach with combinations of different start time and window length parameters to explore the optimal signal trimming interval that would lead to high average and low standard deviation of classification accuracies. Then, we looked for the minimum calibration size, to fulfil a balance between higher accuracy and shorter calibration time, so that the calibration phase of BCI can be eliminated from further training sessions. To improve the explainability of deep learning architectures; cross-correlation analysis and GradCAM methodology were employed to explore the significant input EEG regions that are influential for a correct P300 classification. We observed the consistent attention of a classifier on these regions with saliency maps. Finally, with Grad-CAM methodology, we identified the most significant electrode channels. Their positions on the scalp corresponded to visual processing areas (parietal and occipital lobes). The elimination of less informative electrode channels from the data set were evaluated and the accuracy performance did not degrade much. All the methodologies, explorations and conclusions were performed and validated on two different CNN classifiers, demonstrating the generalizability of the obtained results.
Benzer Tezler
- Efficient human parsing and inpainting using advanced deep learning techniques
Gelı̇şmı̇ş derı̇n öğrenme teknı̇klerı̇nı̇ kullanarak etkı̇n ı̇nsan ayrıştırma ve inpainting
MD IMRAN HOSEN
Doktora
İngilizce
2024
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolBahçeşehir ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
DR. ÖĞR. ÜYESİ TARKAN AYDIN
- Leveraging deep learning for efficient translation of programs between programming languages
Programlama dilleri arasinda verimli kod çevirisi için derin öğrenmeden yararlanma
KHALIL E A ABDULGAWAD
Yüksek Lisans
İngilizce
2025
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİstanbul Aydın ÜniversitesiYazılım Mühendisliği Ana Bilim Dalı
Prof. Dr. ALİ OKATAN
- Statistical feature learning and signal generation for time-series sensor signals
Zaman serisi sensör işaretleri için öznitelik öğrenimi ve işaret üretimi
ERKAN KARAKUŞ
Doktora
İngilizce
2022
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİstanbul Teknik ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
PROF. DR. HATİCE KÖSE
- Digital twin of UAV: İmage processing in different virtual environments
İHA'nın dijital ikizi: Farklı sanal ortamlarda görüntü işleme
SAMET AKÇAY
Yüksek Lisans
İngilizce
2023
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolOstim Teknik ÜniversitesiYazılım Mühendisliği Ana Bilim Dalı
DR. ÖĞR. ÜYESİ İCLAL ÇETİN TAŞ